Assessments of ecosystem functioning are a fundamental ecological challenge and an essential foundation for ecosystem-based management. An understanding of species trophic positions (TP) is essential to characterize food web architecture. However, despite the intuitive nature of the concept, empirically estimating TP is a challenging task due to the complexity of trophic interaction networks. Various alternative methods are proposed to assess TPs, including different approaches to account for the different sources of organic matter at the base of the food web (the “baseline”). However, it is often not clear which methodological approach and which baseline choices are the most reliable. Using an ecosystem-wide assessment of a tropical reef (Marquesas Islands, French Polynesia, with available data for 70 coral reef invertebrate and fish species), we tested whether different commonly used TP estimation methods yield similar results and, if not, whether it is possible to identify the most reliable method. We found significant differences in TP estimates of up to 1.7 TP for the same species, depending on the method and the baseline used. When using bulk stable isotope data, the choice of the baseline significantly impacted TP values. Indeed, while δ¹⁵N values of macroalgae led to consistent TP estimates, those using phytoplankton generated unrealistically low TP estimates. The use of a conventional enrichment factor (i.e. 3.4 ‰) or a “variable” enrichment factor (i.e. according to feeding guilds) also produced clear discrepancies between TP estimates. Regarding the use of different calculation methods, TPs obtained with δ¹⁵N values of source amino acids (compound specifics isotope analysis) were close to those assessed with macroalgae, but evidenced the opposite seasonal pattern, with significantly lower TPs in winter than in summer for the majority of assessed species, with particularly pronounced differences for lower TP species. We use the observed differences to discuss possible drivers of the diverging TP estimates and the potential ecological implications.

The data set for the case study for the method comparison was obtained in Nuku Hiva, the largest of the Marquesas Islands (8°54’S, 140°02’W), French Polynesia. Major local environmental characteristics and the sampling methods were already described in detail in previous studies (Galzin et al., 2016; Fey et al., 2020, 2021). Briefly, the studied area, named the “Baie du controleur”, has relatively strong hydrodynamic conditions, and hosts a marine seafloor dominated by rocky habitats, characterized mainly by steep scree slopes of volcanic rock mixed with patches of soft-bottom habitats, algae groves, coral habitats and caves. The benthic community is composed mainly of algal turf, macroalgae, scattered coral colonies and sponges. Other distinctive features of the studied site include the absence of Acropora spp. corals, which are common across other Polynesian coral reefs, and a mean live coral cover of only ~ 5%. Sampling was realized at two seasons, in August 2016 (austral winter) and March 2017 (summer).

Among the various potential sources of organic matter fuelling the food web (Fey et al., 2020), phytoplankton and macroalgae were overall the most important sources of organic matter in this system (Fey et al., 2021), and were thus considered for the analyses in the present study (phytoplankton: n = 39; macroalgae: n = 71 ; Suppl. Table 1). Among primary consumers, molluscs (gastropods and bivalves) are usually assumed to integrate the baseline with little spatiotemporal fluctuation (Cabana & Rasmussen, 1996; Post, 2002a; Layman et al., 2012). We therefore used the grazing gastropod Mauritia spp. (n= 18) and the filter-feeder oyster Pinctada margaritifera (n= 24) as primary consumers for baseline calculations. Other primary consumers and several secondary-tertiary consumers (invertebrates and fish; n = 3 to 43 depending on the species, for a total of 737 individuals analysed ; Suppl. Table 2) were sampled to assess their TPs. Among secondary-tertiary consumers, we also selected eight mesopredator species expected to be at the top of the local benthic food webs, for a compound specific stable isotope analysis (CS-SIA) (see below). These species were one gastropod, Conus conco, and seven fish: the snappers Lutjanus bohar, L. gibbus, and L. kasmira, the moray-eel Enchelycore pardalis, the scorpionfish Scorpaenopis possi, and the groupers Cephalopholis argus and Epinephelus fasciatus (n= 6 for each species except L. bohar, n = 4).

Invertebrates were collected by handpicking during scuba diving, and fish were collected by spearfishing or using an anesthetic (i.e., eugenol diluted at 10% in alcohol), both in winter and summer. For most animal organisms (total of 70 species), tissues analysed were muscles and, for each taxonomic group, systematically the same location (e.g., dorsal muscle in fish, abductor muscle in bivalves, etc.). For ascidians and sponges, ~ 5–10 g pieces, excluding external theca for ascidians, were taken from each individual.

Animal tissues (muscles or small pieces of organisms, see above) were taken and immediately frozen at −20°C for subsequent analyses. Tissue samples of macro-invertebrates and fish were freeze-dried and grounded to fine powder with a porcelain mortar and pestle. Approximately 1 mg of powder was weighed and encapsulated in tin caps. The bulk δ¹⁵N values were determined using continuous-flow isotope-ratio mass spectrometry with a Flash 2000 elemental analyzer equipped with the Smart EA option (Thermo Scientific, Milan, Italy), coupled with a Delta V Advantage isotope ratio mass spectrometer with a Conflo IV interface (Thermo Scientific, Bremen, Germany) at the Littoral, Environment and Societies Joint Research Unit stable isotope facility (LIENSs) at the University of La Rochelle (France). Calibration was done using reference materials (USGS-61, -62, IAEAN2, –NO–3, − 600 for nitrogen). The analytical precision of the measurements was < 0.15‰ based on analyses of USGS-61 and USGS-62 used as laboratory internal standards.

For δ¹⁵N_AA analyses, samples were prepared by acid hydrolysis followed by derivatization to produce trifluoroacetic amino acid esters (TFAA) using a standard method (Popp et al., 2007). The δ¹⁵N values of the TFAA derivatives of amino acids were analyzed using an isotope ratio mass spectrometer (Delta V Plus, Thermo Scientific, Bremen, Germany) interfaced with a gas chromatograph (Trace GC 1300, Thermo Scientific, Bremen, Germany) through a GC IsoLink combustion furnace, and liquid nitrogen cold trap at the University of Davis (California, USA). Measured isotopic values were corrected relative to known δ¹⁵N values of norleucine, the internal reference material. All samples were analyzed in triplicate. Average standard deviation of triplicate measurements was no greater than ± 1.25 across amino acids (within sample/reference materials) and across samples (within amino acids). Standard deviation of individual amino acids within sample/reference materials was no greater than ± 1.75.